Working with big data

Session 5: Interactive and dynamic graphics

Professor Di Cook

Department of Econometrics and Business Statistics

Why interactive plots

Reasons

  • Engage the reader and allow some choice in what to examine about the data or model.
  • De-clutter the information presented, by only showing some aspects of the data on-demand.
  • Too much information to present in a single plot, so provide multiple plots where the information is linked.
  • Re-scale information to change the focus on-demand.

Interactivity means the user can directly change aspects of the plot using mouse or keyboard controls.

Animation is an alternative to interactivity that keeps control with the developer rather than the reader.

Code 
gp <- gapminder |> 
  filter (year == 2007) |>
  ggplot(aes(x=lifeExp, 
             y=gdpPercap,
             label=country,
             colour=continent)) +
  geom_point() +
  scale_colour_discrete_divergingx(palette = "Zissou 1") +
  scale_y_log10("gdpPercap ('000)",
                breaks = seq(0, 50000, 10000), 
                labels = seq(0, 50, 10)) +
  theme(axis.title = element_text(family="Helvetica"),
        axis.text = element_text(family="Helvetica"),
        legend.title = element_text(family="Helvetica"),
        legend.text = element_text(family="Helvetica")) 
gp + geom_text() +
  ggtitle("Too cluttered")

Keep in mind

  • Interactivity like selection should be precise.
  • Response needs to be fast.
  • Be careful not to inflate plot file size when including in reports or presentations.

Different types of interactivity

Mouse-over

Code 
ggplotly(gp, width=700, height=550) |>
  config(displayModeBar = FALSE)

Notice also the subsetting legend

Mouse-over is very easy to find in many, many software.

Pan/zoom

Code 
ggplotly(gp, width=700, height=550) |>
  config(
         modeBarButtonsToRemove = c('select', 'zoomIn',
                                    'zoomOut', 'autoScale',
                                    'resetScale'))

Selection

Code 
set.seed(802)
gapminder_ts <- gapminder |>
  as_tsibble(index=year, key=c(country, continent)) |>
  sample_n_keys(50)
gphk <- highlight_key(gapminder_ts, ~country)

gpl <- ggplot(gphk, aes(x=year, 
                             y=lifeExp, 
                             group=country)) +
        geom_line() +
        ylab("Life Expectancy") +
        ggtitle("Click on a line to highlight a country") +
        theme(axis.title = element_text(family="Helvetica"),
          axis.text = element_text(family="Helvetica"),
          legend.title = element_text(family="Helvetica"),
          legend.text = element_text(family="Helvetica"),
          title = element_text(family="Helvetica"))

ggpl <- ggplotly(gpl, height = 800, width = 1600) |>
  config(displayModeBar = FALSE)
        
highlight(ggpl)

Linking multiple plots

Code 
gggs <- ggplot(gphk, aes(x=continent, y=country)) +
  geom_point() +
  xlab("") + ylab("") +
  theme(axis.title = element_text(family="Helvetica"),
      axis.text = element_text(family="Helvetica"),
      legend.title = element_text(family="Helvetica"),
      legend.text = element_text(family="Helvetica"),
      title = element_text(family="Helvetica"))

gggspl <- ggplotly(gggs, width=500, height=500) |>
  config(displayModeBar = FALSE) |>
  highlight(on = "plotly_selected", 
            off = "plotly_doubleclick") 

ggpl2 <- ggplotly(gpl, height = 500, width = 1000) |>
  config(displayModeBar = FALSE) |>
  highlight(on = "plotly_selected", 
            off = "plotly_doubleclick") 
  
bscols(gggspl, ggpl2, widths = c(5, 7))

Graphical user interface (GUI) control

Code 
gpl <- ggplot(gphk, aes(x=year, 
                             y=lifeExp, 
                             group=country)) +
        geom_line() +
        ylab("Life Expectancy") + xlab("") +
        theme(axis.title = element_text(family="Helvetica"),
          axis.text = element_text(family="Helvetica"),
          legend.title = element_text(family="Helvetica"),
          legend.text = element_text(family="Helvetica"),
          title = element_text(family="Helvetica"))
ggpl2 <- ggplotly(gpl, height = 500, width = 1000) |>
  config(displayModeBar = FALSE)
  
bscols(widths = c(4, 7),
    filter_select("country", "country", gphk, ~country, multiple = TRUE),
  ggpl2)

GUI elements

GUIs provide explicit control over a small range of interactions.

  • Menu: for a medium number of categories
  • Slider: numeric values or range
  • Checkbox: for a small number of categories


Read more about crosstalk here.

Using sound

Vowel explorer example

Available software

Software list

  • crosstalk: that’s what shiny is based on - we will look into shiny later
  • plotly: interactive javascript graphics (maintained Carson Sievert)
  • leaflet: (RStudio) allows to make interactive maps. Has been picking up users and has developed a stable user base.
  • ggvis: both static and interactive graphics, work on it has stalled … (Wickham)
  • animint2: interactive, linked graphics with ggplot2 syntax (Toby Hocking)
  • rCharts, rbokeh, gridSVG, epivizr, cranvas previous approaches to packages with interactive graphics
  • see also https://r-graph-gallery.com/interactive-charts.html for additional packages (more specific)
  • and CRAN Task View Dynamic Visualizations and Interactive Graphics

plotly

The plotly package in R has two interfaces:

  • plot specification via plotly
  • translating ggplot2 plots and adding interactive elements

Plotly creates interactive plots with javascript.

Code 
plot_ly(data = penguins_std, 
        x = ~fl, 
        y = ~bl, 
        color = ~species, 
        size = 3, 
        width=420, height=300)

plotly

Code 
plot_ly(data = penguins_std, 
        x = ~fl, 
        y = ~bl, 
        color = ~species, 
        size = 3, width=650, 
        height=490, 
        type="scatter", 
        mode="markers")

ggplot2 and plotly

Code 
gg <- ggplot(data=penguins_std, aes(x = fl, 
                                    y = bl, 
                                    colour = species)) +  
  geom_point(alpha=0.5) + 
  geom_smooth(method = "lm", se=F)
ggplotly(gg, width=600, height=490)

Maps

Code 
data(canada.cities, package = "maps")
viz <- ggplot(canada.cities, aes(long, lat)) +
  borders(regions = "canada") +
  coord_equal() +
  geom_point(aes(text = name, size = log2(pop)), 
             colour = "red", alpha = 1/4) +
  theme_map()
Code 
ggplotly(viz)

Not all ggplot2 geoms are supported in plotly, but when they are, they just work out of the box

Modifying plotly

plotly uses elements of crosstalk to provide additional interactivity, such as linked highlighting

Code 
txh_shared <- highlight_key(txhousing, ~year)

p <- ggplot(txh_shared, aes(month, median)) +
   geom_line(aes(group = year)) + 
   geom_smooth(data = txhousing, method = "gam") + 
   scale_x_continuous("", breaks=seq(1, 12, 1),
        labels=c("J", "F", "M", "A", "M", "J", 
                 "J", "A", "S", "O", "N", "D")) +
   scale_y_continuous("Median price ('00,000)", 
                      breaks = seq(0,300000,100000),
                      labels = seq(0,3,1)) +
   facet_wrap(~ city)

gg <- ggplotly(p, height = 750, width = 900) %>%
   plotly::layout(title = "Click on a line to highlight a year")

Code 
highlight(gg)

The power of crosstalk

Code 
tourism_shared <- tourism |>
  as_shared_tsibble(spec = (State / Region) * Purpose)

tourism_feat <- tourism_shared |>
  features(Trips, feat_stl)

p1 <- tourism_shared |>
  ggplot(aes(x = Quarter, y = Trips)) +
  geom_line(aes(group = Region), alpha = 0.5) +
  facet_wrap(~ Purpose, scales = "free_y") +
  theme(axis.title = element_text(family="Helvetica"),
        axis.text = element_text(family="Helvetica"),
        legend.title = element_text(family="Helvetica"),
        legend.text = element_text(family="Helvetica"))
p2 <- tourism_feat |>
  ggplot(aes(x = trend_strength, y = seasonal_strength_year)) +
  geom_point(aes(group = Region)) +
  xlab("trend") + ylab("season") +
  theme(axis.title = element_text(family="Helvetica"),
        axis.text = element_text(family="Helvetica"),
        legend.title = element_text(family="Helvetica"),
        legend.text = element_text(family="Helvetica"),
        plot.title = element_text(family="Helvetica"))

The shared data objects from crosstalk make linking between plots easier!

Code 
subplot(
    ggplotly(p1, tooltip = "Region", width = 1200, height = 600) |>
  config(displayModeBar = FALSE),
    ggplotly(p2, tooltip = "Region", width = 1000, height = 500) |>
  config(displayModeBar = FALSE),
    nrows = 1, widths=c(0.5, 0.5), heights=1) |>
  highlight(dynamic = FALSE)

Case study: mapping

Constructing a choropleth

Code 
# Read the data
# Replace null with 0, for three LGAs
# Convert to long form to join with polygons
# Make the date variables a proper date
# Set NAs to 0, this is a reasonable assumption
covid <- read_csv("data/melb_lga_covid.csv") |>
  mutate(Buloke = as.numeric(ifelse(Buloke == "null", "0", Buloke))) |>
   mutate(Hindmarsh = as.numeric(ifelse(Hindmarsh == "null", "0", Hindmarsh))) |>
   mutate(Towong = as.numeric(ifelse(Towong == "null", "0", Towong))) |>
  pivot_longer(cols = Alpine:Yarriambiack, names_to="NAME", values_to="cases") |>
  mutate(Date = ydm(paste0("2020/",Date))) |>
  mutate(cases=replace_na(cases, 0))
Code 
# Case counts are cumulative, keep only latest
covid <- covid |>
  filter(Date == ymd("2020-10-20"))
Code 
load("data/lga.rda")

covid_tot <- covid |>
  left_join(lga, by=c("NAME" = "lga")) |>
  st_as_sf()

# Make choropleth map, with appropriate colour palette
cm1 <- ggplot(covid_tot) + 
  geom_sf(aes(fill = cases, label = NAME),
    colour="grey80") + 
  scale_fill_distiller("Cases", 
    palette = "PuBuGn",
    direction=1) + 
  theme_map() +
  theme(legend.position="bottom")
cm1

Code 
# Make it interactive
# plotly::ggplotly()

Numerical value of statistic is attached to the respective polygon.



But a problem, especially for Australia is that small geographic, but high population density, areas get lost.

Making the small regions visible (1/2)

Code 
pop <- read_xlsx("data/VIF2019_Population_Service_Ages_LGA_2036.xlsx", sheet=3, skip=13, col_names = FALSE) |>
  select(`...4`, `...22`) |>
  rename(lga = `...4`, pop=`...22`) |>
  filter(lga != "Unincorporated Vic") |> 
  mutate(lga = str_replace(lga, " \\(.+\\)", "")) |>
  mutate(lga = ifelse(lga == "Colac-Otway", "Colac Otway", lga)) 

covid_tot <- covid_tot |>
  left_join(pop, by=c("NAME" = "lga")) 

covid_tot <- covid_tot |>
  mutate(cases_per10k = cases/pop*10000,
         lcases = log10(cases + 1)) 

covid_tot_carto <- covid_tot |> 
  st_transform(3395) |> 
  cartogram_cont("pop") |>
  st_transform("WGS84")   
  
covid_tot_carto <- st_cast(covid_tot_carto, "MULTIPOLYGON") 

cm2 <- ggplot(covid_tot_carto) + 
  geom_sf(aes(fill = cases, label=NAME),
    colour="grey80") + 
  scale_fill_distiller("Cases", palette = "PuBuGn",
                       direction=1) + 
  theme_map() +
  theme(legend.position="bottom")  
cm2

A cartogram expands a geographic are relative to the population in the area.



See more on cartograms here.

A better solution for Australia is needed, though.

Making the small regions visible (2/2)

Code 
# Placement of hexmaps depends on position relative to
# Melbourne central
data(capital_cities)
covid_hexmap <- create_hexmap(
  shp = covid_tot,
  sf_id = "NAME",
  focal_points = capital_cities, verbose = TRUE)

# Hexagons are made with the `fortify_hexagon` function
covid_hexmap_poly <- covid_hexmap |>
  fortify_hexagon(sf_id = "NAME", hex_size = 0.1869) |>
  left_join(covid_tot, by="NAME") # hexmap code removed cases!
cm3 <- ggplot() +
  geom_sf(data=covid_tot, 
          fill = "white", colour = "grey80", size=0.1) +
  geom_polygon(data=covid_hexmap_poly, 
               aes(x=long, y=lat, group=hex_id, 
                   fill = cases, 
                   colour = cases,
                   label=NAME), size=0.2) +
  scale_fill_distiller("Cases", palette = "PuBuGn",
                       direction=1) +
  scale_colour_distiller("Cases", palette = "PuBuGn",
                       direction=1) +
  theme_map() +
  theme(legend.position="bottom")
cm3

Code 
# ggplotly()


Learn more about hexagon tiling that works better for Australia here.

Adding interaction

Code 
cm1 <- cm1 + theme(legend.position = "none")
ggplotly(cm1, width=800, height=600) |>
  config(displayModeBar = FALSE)
Code 
cm2 <- cm2 + theme(legend.position = "none")
ggplotly(cm2, width=800, height=600) |>
  config(displayModeBar = FALSE)
Code 
cm3 <- cm3 + theme(legend.position = "none")
ggplotly(cm3, width=800, height=600) |>
  config(displayModeBar = FALSE)

Animation

Explanation

  • gganimate (Lin-Pederson) allows to make and save animations: gganimate cheat sheet
  • Animations are different from interactive graphics in that the viewer does not have any control
  • useful for different important stages of a visualization (e.g. time) and to keep track of how different visualizations are related
  • can also be used in talks

An example animation

Code 
ggplot(gapminder, aes(gdpPercap, lifeExp, size = pop, colour = country)) +
  geom_point(alpha = 0.7) +
  scale_colour_manual(values = country_colors) +
  scale_size("Population size", range = c(2, 12), breaks=c(1*10^8, 2*10^8, 5*10^8, 10^9, 2*20^9)) +
  scale_x_log10() +
  guides(colour = "none") +
  facet_wrap(~continent) +
  theme(legend.position = "bottom") +
  # Here comes the gganimate specific bits
  labs(title = 'Year: {frame_time}', x = 'GDP per capita', y = 'life expectancy') +
  gganimate::transition_time(year) +
  gganimate::ease_aes('linear')

Countries are colored manually by country_colors (hue shows continent, saturation is individual country)

gganimate

  1. Start with a ggplot2 specification
  2. Add layers with graphical primitives (geoms)
  3. Add formatting specifications
  4. Add animation specifications

A simple example

  1. Start by passing the data to ggplot
Code 
ggplot(economics) #<<

Thanks to Mitch O’Hara Wild for the example

A simple example

  1. add the mapping
Code 
ggplot(economics) +
  aes(date, unemploy) #<<

Thanks to Mitch O’Hara Wild for the example

A simple example

  1. Add a graphical primitive, let’s do a line
Code 
ggplot(economics) +
  aes(date, unemploy) +
  geom_line() #<<

Thanks to Mitch O’Hara Wild for the example

A simple example

  1. Just one extra line turns this into an animation!
Code 
ggplot(economics) +
  aes(date, unemploy) +
  geom_line() +
  transition_reveal(date) #<<

Thanks to Mitch O’Hara Wild for the example

A not-so-simple example

Using the the datasaurus dozen, again, we first pass in the dataset to ggplot

Code 
ggplot(datasaurus_dozen)#<<

A not-so-simple example

For each dataset we have x and y values, in addition we can map dataset to color

Code 
ggplot(datasaurus_dozen) +
  aes(x, y, color=dataset) #<<

A not-so-simple example

Trying a simple scatter plot first, but there is too much information

Code 
ggplot(datasaurus_dozen) +
  aes(x, y, color=dataset) +
  geom_point() + #<<
  xlim(c(0,100)) + ylim(c(0,100)) +
  coord_equal()

A not-so-simple example

We can use facets to split up by dataset, revealing the different distributions

Code 
ggplot(datasaurus_dozen) +
  aes(x, y, color=dataset) +
  geom_point() +
  facet_wrap(~dataset) + #<<
  xlim(c(0,100)) + ylim(c(0,100)) +
  coord_equal() +
  theme(legend.position = "none")

A not-so-simple example

We can just as easily turn it into an animation, transitioning between dataset states!

Code 
ggplot(datasaurus_dozen) +
  aes(x, y) +
  geom_point() +
  xlim(c(0,100)) + ylim(c(0,100)) +
  coord_equal() +
  transition_states(dataset, 2, 3) + #<<
  labs(title = "Dataset: {closest_state}") #<<

Controlling an animation

We control plot movement with (a grammar of animation):

  • Transitions: transition_*() define how the data should be spread out and how it relates to itself across time.
  • Views: view_*() defines how the positional scales should change along the animation.
  • Shadows: shadow_*() defines how data from other points in time should be presented in the given point in time.
  • Entrances/Exits: enter_*() and exit_*() define how new data should appear and how old data should disappear during the course of the animation.
  • Easing: ease_aes() defines how different aesthetics should be eased during transitions.

Resources

  • plotly
  • gganimate
  • crosstalk

Creative Commons License This work is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License.